MARBLE
收藏资源简介:
MARBLE是一个用于评估多模态语言模型(MLLMs)在复杂多模态问题和环境中的推理和规划能力的高难度基准。数据集包含两个挑战性任务:M-Portal和M-Cube,它们要求在空间、视觉和物理约束下制定和理解多步骤计划。M-Portal任务涉及规划、空间推理和多模态集成,灵感来源于游戏Portal 2。M-Cube任务要求将六块拼图式的拼块组装成目标立方体形状,灵感来源于Happy Cube拼图。每个任务都包含两个不同难度的子任务。数据集共有1536个样本,旨在推动下一代模型在多步多模态推理和规划能力方面的发展。
MARBLE is a challenging benchmark for evaluating the reasoning and planning capabilities of Multimodal Language Models (MLLMs) in complex multimodal problems and environments. The dataset comprises two demanding tasks: M-Portal and M-Cube, which require formulating and understanding multi-step plans under spatial, visual, and physical constraints. The M-Portal task involves planning, spatial reasoning, and multimodal integration, inspired by the video game Portal 2. The M-Cube task requires assembling six puzzle pieces into a target cube shape, with inspiration drawn from the Happy Cube puzzle. Each task includes two subtasks of differing difficulty levels. The dataset consists of 1,536 total samples, designed to drive the advancement of multi-step multimodal reasoning and planning capabilities for next-generation models.
MARBLE: 多模态空间推理与规划的硬基准
数据集概述
- 包含2个子数据集:M-Portal和M-CUBE
- 每个子数据集包含2个子任务
- 设计目的:检验多模态语言模型(MLLMs)在复杂多模态问题中的逐步推理能力
M-Portal数据集
基本信息
- 数据量:1,024个视觉问答任务
- 数据来源:16个精心策划的《Portal 2》游戏地图
- 子任务:
- Plan-correctness(计划正确性验证):512个问题
- Fill-the-blanks(填补缺失步骤):512个问题
模型表现
| 模型 | Plan-correctness (F1 %) | Fill-the-blanks (Acc %) |
|---|---|---|
| GPT-o3 | 6.6 | 17.6 |
| Gemini-2.5-pro | 4.7 | 16.1 |
| DeepSeek-R1-0528* | 0.0 | 8.4 |
| Claude-3.7-Sonnet | 6.3 | 6.8 |
| DeepSeek-R1* | 6.1 | 5.5 |
| Seed1.5-VL | 7.6 | 3.5 |
| GPT-o4-mini | 0.0 | 3.1 |
| GPT-4o | 6.5 | 0.4 |
| Llama-4-Scout | 6.5 | 0.2 |
| Qwen2.5-VL-72B | 6.6 | 0.2 |
| InternVL3-78B | 6.4 | 0.0 |
| Qwen3-235B-A22B* | 0.0 | 0.0 |
| Random | 6.1 | 3e-3 |
M-CUBE数据集
基本信息
- 任务类型:3D空间推理
- 核心任务:将6个拼图块组装成立方体
- 数据量:
- CUBE(完整难度):1,000个示例
- CUBE-easy(简化版):1,000个示例
模型表现
| 模型 | CUBE (Acc %) | CUBE-easy (Acc %) |
|---|---|---|
| GPT-o3 | 0.0 | 72.0 |
| GPT-o4-mini | 0.0 | 16.0 |
| DeepSeek-R1* | 0.0 | 14.0 |
| Gemini-2.5-pro | 0.0 | 11.0 |
| DeepSeek-R1-0528* | 0.0 | 8.0 |
| Claude-3.7-Sonnet | 0.0 | 7.4 |
| InternVL3-78B | 0.0 | 2.8 |
| Seed1.5-VL | 0.0 | 2.0 |
| GPT-4o | 0.0 | 2.0 |
| Qwen2.5-VL-72B | 0.0 | 2.0 |
| Llama-4-Scout | 0.0 | 1.6 |
| Qwen3-235B-A22B* | 0.0 | 0.3 |
| Random | 1e-5 | 3.1 |
引用信息
bibtex @article{jiang2025marble, title={MARBLE: A Hard Benchmark for Multimodal Spatial Reasoning and Planning}, author={Jiang, Yulun and Chai, Yekun and Brbic, Maria and Moor, Michael}, journal={arXiv preprint arXiv:2506.22992}, year={2025}, url={http://arxiv.org/abs/2506.22992} }




